Our names are: Shan Liang, Wei Jiang and Wenju Liu. We works in National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences. Email: {sliang, wjiang, lwj}@nlpr.ia.ac.cn. We first find the continuous voiced segments by tracked pitch countour from mixture. We then get a rough estimate of noise from unvoiced segments. The T-F unit power higher the estimated power over a critical band is considered as speech dominant unit. The methods [2] and [3] are also used to revise the IBM estimation to some extent. This results in a rough estimate of IBM. Then, we use the method [4] to obtain a more accurte estimate of noise according the estimated IBM. We also adjust the IBM estimate with the update of noise estimate. The algorithm is running on a computer with Intel(R) Core(TM) i3-2310M 2.10GHz CPU. The average running time is 39.1(s). And the time for each stereo sentence is as follows: test/test_Ca1_Ce_A_mix.wav 39.608654 test/test_Ca1_Ce_B_mix.wav 39.405852 test/test_Ca1_Co_A_mix.wav 40.295059 test/test_Ca1_Co_B_mix.wav 39.171851 test/test_Ca2_Ce_A_mix.wav 39.671054 test/test_Ca2_Ce_B_mix.wav 39.733455 test/test_Ca2_Co_A_mix.wav 40.029857 test/test_Ca2_Co_B_mix.wav 42.120270 test/test_Sq1_Ce_A_mix.wav 40.872262 test/test_Sq1_Ce_B_mix.wav 37.268639 test/test_Sq1_Co_A_mix.wav 38.579047 test/test_Sq1_Co_B_mix.wav 38.329445 test/test_Sq2_Ce_A_mix.wav 38.937850 test/test_Sq2_Ce_B_mix.wav 39.062650 test/test_Sq2_Co_A_mix.wav 38.469847 test/test_Sq2_Co_B_mix.wav 37.424640 test/test_Su1_Ce_A_mix.wav 37.549441 test/test_Su1_Ce_B_mix.wav 36.941037 test/test_Su2_Ce_A_mix.wav 39.203052 test/test_Su2_Ce_B_mix.wav 39.359052 dev/dev_Ca1_Ce_A_mix.wav 41.090664 dev/dev_Ca1_Ce_B_mix.wav 40.997062 dev/dev_Ca1_Co_A_mix.wav 38.969049 dev/dev_Ca1_Co_B_mix.wav 38.547847 dev/dev_Sq1_Ce_A_mix.wav 37.346640 dev/dev_Sq1_Ce_B_mix.wav 38.329446 dev/dev_Sq1_Co_A_mix.wav 36.597835 dev/dev_Sq1_Co_B_mix.wav 37.643041 dev/dev_Su1_Ce_A_mix.wav 39.764655 dev/dev_Su1_Ce_B_mix.wav 40.357458 Key References include: [1]Ma, W., Yu, M., Xin, J., Osher, S.: Reducing musical noise in blind source separation by time-domain sparse filters and split bregman method. Interspeech 2010, pp. 402-405. [2]Yilmaz, O., Rickard, S.: Blind separation of speech mixtures via time-frequency masking. IEEE Trans. Signal Processing. vol. 52(7), pp. 1830--1847 (2004). [3]Shimamura, T., Yamauchi, J.: Non-stationary noise estimation utilizing harmonic structure for spectral subtraction. Conference record of the thirty-eighth asilomar conference on signals, systems and computers. vol. 2, pp. 2305--2609 (2004). [4]Rangachari, S., Loizou, P. C.: A noise-estimation algorithm for highly non-stationary environments. IEEE Trans. Speech Commun. vol. 48(2), pp. 220--231 (2006).